In [0]:
from fastai.vision import *
from fastai import *

Choose an appropriate name for your labeled images. You can run these steps multiple times to create different labels.

In [14]:
folder = 'Vader'
from google.colab import drive
drive.mount('/content/gdrive', force_remount=True)
root_dir = "/content/gdrive/My Drive/"
base_dir = root_dir + 'fastai-v3/data'
file = 'urls_vader.csv'

path = Path(base_dir)
dest = path/folder
dest.mkdir(parents=True, exist_ok=True)

Mounted at /content/gdrive
In [15]:
folder = 'Kylo'
drive.mount('/content/gdrive', force_remount=True)
root_dir = "/content/gdrive/My Drive/"
base_dir = root_dir + 'fastai-v3/data'
file = 'urls_kylo.csv'

path = Path(base_dir)
dest = path/folder
dest.mkdir(parents=True, exist_ok=True)

Mounted at /content/gdrive

You will need to run this cell once per each category.

In [17]:
path.ls()
Out[17]:
[PosixPath('/content/gdrive/My Drive/fastai-v3/data/Vader'),
 PosixPath('/content/gdrive/My Drive/fastai-v3/data/Kylo'),
 PosixPath('/content/gdrive/My Drive/fastai-v3/data/vader'),
 PosixPath('/content/gdrive/My Drive/fastai-v3/data/kylo'),
 PosixPath('/content/gdrive/My Drive/fastai-v3/data/models'),
 PosixPath('/content/gdrive/My Drive/fastai-v3/data/cleaned.csv')]

Download images

In [0]:
classes = ['vader','kylo']
In [0]:
help(download_images)

Help on function download_images in module fastai.vision.data:

download_images(urls:Collection[str], dest:Union[pathlib.Path, str], max_pics:int=1000, max_workers:int=8, timeout=4)
    Download images listed in text file `urls` to path `dest`, at most `max_pics`
In [0]:
download_images(path/folder/file, dest, max_pics=200)
100.00% [200/200 00:25<00:00]
In [0]:
download_images(path/file, dest, max_pics=20, max_workers=0)

Then we can remove any images that can't be opened:

In [19]:
for c in classes:
    print(c)
    verify_images(path/c, delete=True, max_size=500)

vader
100% [0/0] 
/usr/local/lib/python3.6/dist-packages/fastprogress/fastprogress.py:102: UserWarning: Your generator is empty.
  warn("Your generator is empty.")

kylo
100% [0/0]

View data

In [0]:
np.random.seed(42)
data = ImageDataBunch.from_folder(path, train=".", valid_pct=0.2,
        ds_tfms=get_transforms(), size=224, num_workers=4).normalize(imagenet_stats)
In [0]:
#If you already cleaned your data, run this cell instead of the one before
np.random.seed(42)
data = ImageDataBunch.from_csv(path, folder=".", valid_pct=0.2, csv_labels='cleaned.csv',
         ds_tfms=get_transforms(), size=224, num_workers=4).normalize(imagenet_stats)

Good! Let's take a look at some of our pictures then.

In [21]:
data.classes
Out[21]:
['Kylo', 'Vader']
In [0]:
data.show_batch(rows=3, figsize=(7,8))
In [22]:
data.classes, data.c, len(data.train_ds), len(data.valid_ds)
Out[22]:
(['Kylo', 'Vader'], 2, 286, 71)

Train model

In [0]:
learn = cnn_learner(data, models.resnet50, metrics=error_rate)
In [24]:
learn.fit_one_cycle(4)
epoch train_loss valid_loss error_rate time
0 0.812963 0.329234 0.154930 03:40
1 0.510841 0.137943 0.042254 03:36
2 0.349132 0.152274 0.070423 03:34
3 0.284493 0.131682 0.070423 03:34
In [0]:
learn.save('stage-1')
In [0]:
learn.unfreeze()
In [27]:
learn.lr_find()

LR Finder is complete, type {learner_name}.recorder.plot() to see the graph.
In [28]:
learn.recorder.plot()
In [29]:
learn.fit_one_cycle(2, max_lr=slice(3e-5,3e-4))
epoch train_loss valid_loss error_rate time
0 0.093106 0.063346 0.028169 04:30
1 0.071501 0.067660 0.028169 04:30
In [0]:
learn.save('stage-2')

Interpretation

In [0]:
learn.load('stage-2');
In [0]:
interp = ClassificationInterpretation.from_learner(learn)
In [33]:
interp.plot_confusion_matrix()

Exporting

In [0]:
learn.export()
In [0]:
defaults.device = torch.device('cpu')
In [39]:
img = open_image(path/'vader-2.jpg')
img
Out[39]:

We create our Learner in production enviromnent like this, jsut make sure that path contains the file 'export.pkl' from before.

In [0]:
learn = load_learner(path)
In [41]:
pred_class,pred_idx,outputs = learn.predict(img)
pred_class
Out[41]:
Category Vader
In [49]:
img = open_image(path/'kylo-2.jpg')
img
Out[49]:
In [50]:
pred_class,pred_idx,outputs = learn.predict(img)
pred_class
Out[50]:
Category Kylo
In [53]:
img = open_image(path/'vader-7.jpg')
img
Out[53]:
In [54]:
pred_class,pred_idx,outputs = learn.predict(img)
pred_class
Out[54]:
Category Vader
In [56]:
img = open_image(path/'vader-8.jpg')
img
Out[56]:
In [58]:
pred_class,pred_idx,outputs = learn.predict(img)
pred_class
Out[58]:
Category Kylo
In [59]:
img = open_image(path/'kylo-8.jpg')
img
Out[59]:
In [60]:
pred_class,pred_idx,outputs = learn.predict(img)
pred_class
Out[60]:
Category Kylo
In [61]:
img = open_image(path/'kylo-1.jpg')
img
Out[61]:
In [62]:
pred_class,pred_idx,outputs = learn.predict(img)
pred_class
Out[62]:
Category Kylo
In [63]:
img = open_image(path/'vader-3.jpg')
img
Out[63]:
In [64]:
pred_class,pred_idx,outputs = learn.predict(img)
pred_class
Out[64]:
Category Vader